ARTICLESPUBLISHED ONLINE: 21 AUGUST 2017 | DOI: 10.1038/NGEO3012

Anatomy of a meltwater drainage system beneaththe ancestral East Antarctic ice sheetLauren M. Simkins1*, John B. Anderson1, Sarah L. Greenwood2, Helge M. Gonnermann1,Lindsay O. Prothro1, Anna RuthW. Halberstadt1,3, Leigh A. Stearns4, David Pollard5

and Robert M. DeConto3

Subglacial hydrology is critical to understand the behaviour of ice sheets, yet active meltwater drainage beneath contemporaryice sheets is rarely accessible to direct observation. Using geophysical and sedimentological data from the deglaciated westernRoss Sea, we identify a palaeo-subglacial hydrological system active beneath an area formerly covered by the East Antarcticice sheet. A long channel network repeatedly delivered meltwater to an ice stream grounding line and was a persistentpathway for episodic meltwater drainage events. Embayments within grounding-line landforms coincide with the locationof subglacial channels, marking reduced sedimentation and restricted landform growth. Consequently, channelized drainageat the grounding line influenced the degree to which these landforms could provide stability feedbacks to the ice stream.The channel network was connected to upstream subglacial lakes in an area of geologically recent rifting and volcanism,where elevated heat flux would have produced su�cient basal melting to fill the lakes over decades to several centuries; thistimescale is consistent with our estimates of the frequency of drainage events at the retreating grounding line. Based on thesedata, we hypothesize that ice stream dynamics in this region were sensitive to the underlying hydrological system.

Subglacial processes influence the behaviour of ice sheets andtheir grounding lines, themost downstream location ice sheetsare in contact with the underlying bed. In particular, meltwa-

ter beneath ice sheets is associated with the onset of fast-flowingice streams1, shear margins that separate fast from slow ice flow2,and enhanced deformation of subglacial sediments3,4. Meltwaterstored within subglacial lakes5–7 can drain over periods of monthsto several years8–11 due to changes in hydraulic gradient that areprobably triggered by ice thinning and grounding-line retreat10,11.Downstream of draining subglacial lakes, periods of fluctuatingand accelerated ice flow have been interpreted to result from dis-tributed water flow10,12–14. In contrast, decelerated ice flow has beenattributed to lowered basal water pressures, as a consequence ofchannelized meltwater drainage15–17. The distribution and move-ment of subglacial meltwater, therefore, must be well understood toassess changes in ice flow, yet models and theory have outpaced ourknowledge of subglacial hydrology based on direct observations.

A key question regards the influence of subglacial meltwaterdrainage on grounding-line dynamics. The termination of asubglacial channel beneath the contemporary Whillans Ice Stream(Fig. 1a) coincides with a grounding-line embayment, where thegrounding line is located several kilometres further inland from theadjacent grounding line and sediment erosion within the channeland water mixing may alter grounding-line behaviour18,19. Whethera causal relationship exists between channelized drainage andgrounding-line embayments is unclear; however, the possibilitythat channelized drainage influences grounding-line position andsediment accumulation, which can reduce the ice thickness neededto remain in contact with the bed and even facilitate ice advance20,21,should be explored. Furthermore, subglacial channels draining at

the grounding line can release buoyant meltwater plumes thatthermally erode channels into the base of ice shelves22,23. Onesuch example suggests basal melt rates of >15myr−1 withinan actively forming ice shelf channel that is connected to ahypothesized subglacial channel at the grounding line24. Althoughthese observations demonstrate that subglacial channels drain atgrounding lines and can cause ice shelf melting, their impact ongrounding-line dynamics remains tenuous.

The geologic record can provide broader spatial and temporalperspectives on subglacial hydrology. Numerous palaeo-subglacialchannels incised into bedrock are exposed on the Antarcticcontinental shelf25–29 (Fig. 1a), but the timing of their incision andmeltwater occupation is not well constrained. Surficial sediment-based subglacial channels on the Antarctic continental shelf aretemporally constrained30,31, yet they have not thus far been linkedto former grounding lines. Using geophysical and sedimentologicaldata, we provide the first evidence of a subglacial hydrologicalsystem that was active during the post-Last Glacial Maximum(LGM) deglaciation and connected explicitly to grounding-linepositions of a former ice stream in the western Ross Sea (Fig. 1a).

Channelized drainage at palaeo-grounding-line positionsMultibeam bathymetry data with sub-metre vertical resolutionreveal a subglacial hydrological system that spans over 200 kmin distance, with cross-cutting relationships between glacial land-forms that record post-LGM ice flow and retreat of an ice streamsourced from the East Antarctic ice sheet (EAIS) (Fig. 1b). For aregional-scale reconstruction of ice flow and retreat in the Ross Sea,readers are referred to ref. 32. Grounding-line landforms, includ-ing recessional moraines and grounding-zone wedges, represent

1Department of Earth, Environment and Planetary Sciences, Rice University, Houston, Texas 77005, USA. 2Department of Geological Sciences, StockholmUniversity, Stockholm 10691, Sweden. 3Department of Geosciences, University of Massachusetts, Amherst, Massachusetts 01003, USA. 4Department ofGeology, University of Kansas, Lawrence, Kansas 66045, USA. 5Earth and Environmental Systems Institute, Pennsylvania State University, University Park,Pennsylvania 16802, USA. *e-mail: lsimkins@rice.edu

NATURE GEOSCIENCE | VOL 10 | SEPTEMBER 2017 | www.nature.com/naturegeoscience 691

© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.

ARTICLES NATURE GEOSCIENCE DOI: 10.1038/NGEO3012

N

180170160

74

76

Latitude (° S)

Longitude (° E)

78Meltwater deposits

Water depth300

900

Ice elevation100mm

0Volcanic features

Ross IsMt. Erebus

Ross BankCentralBasin

S. DryglaskiTrough

McMurdoSound

PennellBank

Glomar-Challenger

Trough

Franklin Is

MawsonBank

CraryBank

JoidesTroughN. Dryglaski

Trough

Coulman Is

PennellTrough

LGM grounding line

EasternRoss Sea

WesternRoss Sea

MB

ASE

PD

Western Ross Sea

Weddell Sea

150 km

a

1b

Groundedice sheet Floating

ice shelf

125634

∗ Not to scale

4−6: GZWs (2−20 mamp., 50−1,200 m long,500−4,500 m spacing)

Grounding-lineembayments (E)

1−3: Recessionalmoraines (0.5−5 m amp.,20−160 m long,150−1,700 m spacing)

Sea level

Subglaciallineations

Meltwater plumePalaeo-subglacial channels

Groundingline

TopsetForeset

E E

Volcanic featuresSubglacial channel network Subglacial lakes

25 km

Waterdepth

300m

900

2a

3a

4c−d

Upstreamsegment

Middlesegment

Downstreamsegment

Ice flowdirection

Retreatdirection

Activesubglacialchannel

LGM glacialunconformity

b

c

Figure 1 | Western Ross Sea bathymetry and landforms. a, Bathymetryfrom ref. 49. Ice surface elevation from ref. 50. Inset shows Mertz Trough(blue circle), Whillans (green circle) and MacAyeal (purple circle) icestreams, Pine Island Glacier (orange circle), and palaeo-subglacial channelson the continental shelf. ASE, Amundsen Sea Embayment; MB, MargueriteBay; PD, Palmer Deep. b, Overview of the mapped subglacial hydrologicalsystem. c, Schematic of the grounding-line environment and landforms.Moraines formed at times 1–3, followed by retreat to new positions attimes 4–6 marked by grounding-zone wedges (GZWs). Channelizeddrainage bisected a moraine at time 2 and produced embayments at times4 and 6.

periods of grounding-line position stability on the continental shelf(Fig. 1c). Recessional moraines have symmetric geometries andform by sediment deformation and/or deposition at groundinglines33,34. Grounding-zone wedges form by sediment delivery fromthe subglacial environment to marine-based grounding lines andare characterized by asymmetric geometries that result from topsetaggradation and foreset progradation20,33,35,36. Across the subglacial

hydrological system, recessional moraines range in amplitude from0.5–5m and grounding-zone wedges are 2–20m in amplitude(Fig. 1c). We separate the hydrological system into three segmentsfor the purposes of discussion (Fig. 1b): the downstream segmentin southern Joides Trough (Fig. 2), the middle segment south ofCrary Bank (Fig. 3), and the upstream segment containing a seriesof subglacial lakes (Fig. 4).

The downstream segment contains a long primary channel,several smaller channels, and a suite of grounding-line landforms(Fig. 2a). The primary channel is locally buried by recessionalmoraines, which indicates that it initially formed as a subglacialchannel prior to southward grounding-line retreat across the area(Fig. 2b). A core collected at the channel base (KC28; Fig. 2a)supports channel incision into till deposited during or following theLGM. The channel bisects several recessional moraines (Fig. 2c),implying meltwater discharge events either caused local incision orrestricted the growth of the moraines when the grounding line waspositioned at those locations.

Following southward retreat from the positions of the moraines,the grounding line is expressed as a composite grounding-zonewedge (GZW1; 20m in amplitude) comprised of two stackedgrounding-zone wedges with multiple embayments∼1–7 km widealong the topset–foreset break (Fig. 2a). The embayments indicatevariability in grounding-line processes that resulted in a sinuousgrounding line, in contrast to the earlier formed linear recessionalmoraines that predominantly overprint the channel to the north.Channels emanate from each embayment, while a 2.5-m-deep sub-glacial channel occurs on the topset of GZW1 (Fig. 2c,d). The depthof channel incision on the topset of GZW1 is far less than the thick-ness of till (20m) that composes the grounding-zone wedge andleads us to suggest that this channel segment was incised either dur-ing or following the active growth of GZW1. However, local burialof the channel by till indicates grounding-zone wedge constructionwas not complete when the channel was active because sedimentswere later mobilized across segments of the channel (Fig. 2c).

The coincident occurrence of channels and embayments, and theconnection of the topset channel to the larger embayment, impliesthat the embayments were locations of channelized drainage at thegrounding line, comparable in scale to the contemporary Whillansgrounding-line embayment18 and perhaps similar to embayments(termed ‘breach point’ and ‘indentation’) within grounding-zonewedges in Mertz Trough37. The preservation of older recessionalmoraines at the base of the larger embayment is evidence that theembayments formed by restricted sediment deposition, not erosion,at the grounding line. Contrastingly, we observe grounding-zonewedge lobes that prograded over moraines where channels donot occur (Fig. 2c), indicating ice advance was facilitated duringthe northward expansion of the grounding-zone wedge. Thus,embayments mark sites where local ice stability, usually enhancedby grounding-zone wedge construction, was reduced, contributingto spatial variability in grounding-line sensitivity to flotation.Instability could have been further enhanced by melting of iceseaward of the grounding line by buoyant meltwater plumes. Theapparent bisection of several recessional moraines to the north ofGZW1 (Fig. 2c) may be similar, but smaller scale, embayments.Additionally, the superposition of embayments and previouslyformed subglacial channels indicate preferential drainage pathwaysspanning prolonged periods of grounding-line recession andlandform growth.

After ice decoupled from GZW1, the grounding line retreatedsouthwards to another composite grounding-zone wedge (GZW2)with two more grounding-line embayments (Fig. 2a). Thegrounding line then retreated 25 km to the south, where the retreatdirection shifted from south to west across the middle segment ofthe subglacial hydrological system (Figs 1b and 3a). A suite of reces-sionalmoraines and small (<10m in amplitude) lineated grounding

692

© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.

NATURE GEOSCIENCE | VOL 10 | SEPTEMBER 2017 | www.nature.com/naturegeoscience

NATURE GEOSCIENCE DOI: 10.1038/NGEO3012 ARTICLES

Topsetchannelw

w′

Channel bisectsrecessional moraines

Restrictedprogradation

within embayment

GZWprogradation

over moraines

GZWprogradation

over morainesand channel

Water depth580

640

0.0 0.2 0.4 0.6 0.8 1.0598

597

596

595

Wat

er d

epth

(m)

w w′

Distance (km)

GZW1sediment

GZW1sediment

Water depth570

615

Water depth560

700

m m

m

NBP1502Acores

20 km

Downstream segmenta b c

Depth range: 2.5−22 m Width range: 125−750 mArea range: 155−8,250 m2

Field ofrecessionalmoraines

Channels

ChannelGZW1

GZW1

Embayment

Embayments

EmbaymentEmbayments

KC04

KC05

KC34

KC28

Fig. 2b

N

Ice flowdirection

Retreatdirection

Fig. 3aGZW1

Fig. 2c

Recessionalmoraines exposedon flank of GZW1

GZW2

Channel onGZW1 topset

d

Figure 2 | Downstream hydrological segment. a, One long primary channel, several isolated channels, and grounding-line embayments within compositegrounding-zone wedges 1 (GZW1) and 2 (GZW2) in southern Joides Trough (location indicated in Fig. 1b). The topset–foreset breaks of GZW1 and GZW2are outlined in white. b, Channel locally buried by recessional moraines. c, A subglacial channel on the topset of GZW1 terminates at the location of thelarger grounding-line embayment, flanked by GZW lobes that prograded over recessional moraines. d, Profile of shallow channel incised into the topset ofGZW1 (location in c).

0.0 0.4 0.8 1.2 1.6

Distance (km)

690

685

680

675

x′xWat

er d

epth

(m)

Water depth600

800

m

m

NBP1502A cores

Fig. 2aMiddle segmentDepth range: 3.5−12 m Width range: 150−500 mArea range: 260−3,000 m2

Subglaciallineationsoverprintchannel

GZWs bisectedby channel

Volcanic features

Basementrelief low

Basementrelief low

Basementhigh

GZW onbasement high

KC42a

Areas void ofglacial landforms

KC41KC36

Recessionalmorainesoverprintchannel

Fig. 3b

20 km

Franklin Island

Channel

Ice flowdirection

Retreatdirection

Seamounts

Water depth670

715

Recessionalmoraines

GZWsx

x′Embayments

N

bc

Figure 3 | Middle hydrological segment. a, A single channel is locally buried by recessional moraines and subglacial lineations (location in Fig. 1b).Upstream of the channel to the northwest, smooth areas of the seafloor lack expressions of grounded ice, interpreted as subglacial lakes that deliveredmeltwater to the channel network through basement relief lows. b, Grounding-zone wedges (GZWs) were bisected by the channel as the grounding-lineretreated to the west, some of which display small embayments. c, Profile across the channel (location in b).

NATURE GEOSCIENCE | VOL 10 | SEPTEMBER 2017 | www.nature.com/naturegeoscience

© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.

693

ARTICLES NATURE GEOSCIENCE DOI: 10.1038/NGEO3012

0.40.350.09

0.08

0.07

0.065

0.025

0.002

0.22

0.8

10 km10 km

Artifact

y

y′

z

z′

1

2

3

a b

0 5 10 15 20 25 30 35 40 45

Distance (km)

720700680660

Wat

er d

epth

(m)

z z′

Ice flow directionRetreat direction

Water depth600

810

m

Subglacial lakes

40 m

y y′

y y′

Acoustically transparent drape

Strong basal reflection

Thin till unit(KC42)

Acoustically laminatedslope drape

Basin filling, acousticallylaminated reflections

Line turn

NE

850 m

Subglacial lake deposits

GZW

Meltwater fandeposits

Open-marinedrape

Basement

Open-marine depositsSubglacial lake deposits

GZW Basement

Melwater deposits

Strike view Dip view

40 m

850 m

12 3

Middlehydrological

segment

N

c

d

e

Figure 4 | Subglacial lakes in upstream segment. a, Grounding-linelandforms surround basins interpreted as subglacial lakes (location inFig. 1b). b, Same as a, but showing identified subglacial lakes (volumecapacities in km3). White arrows show lows in basement relief throughwhich the lakes probably drained. Dashed line denotes the area(1,300 km2) used to estimate local meltwater production. c, Acousticprofile y–y′ across a small basin shows a strong basal reflection, twolaminated units, and an acoustically transparent unit (location in a).d, Interpreted acoustic profile shown in c. e, Profile z–z′ across the lakes(location in a). Basement highs labelled 1–3.

zone wedges record westward grounding-line retreat, roughlyperpendicular to the channel (Fig. 3a). Recessional moraines andsubglacial lineations locally overprint the channel (Fig. 3a), whichmust therefore have been inactive while sediment was mobilizedacross the channel. In contrast, the channel bisects—or facilitatedembayments within—grounding-zone wedges north of the volcanicseamounts (Fig. 3b,c). These variable relationships between thechannel and other landforms again indicate episodic meltwaterdrainage via the channel network during grounding-line retreat.

The spatial association of subglacial channels and formergrounding-line positions indicates that the channel network asa whole was a persistent feature during numerous retreat eventsand periods of ice flow reorganization. Grounding-line landformsthat both overprint and were bisected by the channel are orientedperpendicular to the channel, while subglacial lineations are broadly

parallel to it. This indicates that ice flow direction and the trajectoryof the retreating grounding line maintained a distinct relationshipto the established channel network (Fig. 1b), and perhaps indicatessensitivity of ice flow pathways to channelized subglacial drainage,yet we cannot constrain a mechanism for this relationship based onour observations.

Frequency and nature of channelized discharge eventsChannelized drainage was predominantly active where grounding-line positions are expressed by grounding-zone wedges andlargely dormant during the formation of recessional moraines(Figs 2 and 3). This implies that the frequency of meltwater dis-charge events typically exceeded the duration of recessionalmoraineformation and, at most, corresponded to the timing of grounding-zone wedge formation. Grounding-zone wedges several tens to hun-dreds of metres in amplitude form over centuries to millennia36,38.However, using the largest grounding-zone wedge in our study area(GZW1; Figs 1c and 2a,b) and a range of sediment fluxes (Methods),the maximum construction time ranges between 80 and 500 years.Smaller grounding-line landforms bisected by the channel probablyformed over years to decades, but we conservatively estimate thefrequency of discharge events to decades to several centuries.

We use the smallest channel geometry (Fig. 2d) to providean estimate of palaeo-meltwater flow conditions, based on ahydropotential surface obtained from palaeo-ice surface and bedelevations (Methods and Supplementary Fig. 1). Calculated bank-full flow velocity is about 1m s−1 with a discharge of 150m3 s−1.Given uncertainties in assumptions and modelled parametersto calculate flow properties, these should be treated only asorder-of-magnitude estimates. However, the calculated dischargeis comparable to peak discharges from contemporary Antarcticsubglacial lake drainage events9,13,39 and meltwater outflow at theSiple Coast grounding line40.

Sedimentological data help us further characterize the natureof drainage at palaeo-grounding-line positions. Relatively sorted,terrigenous fine silts with little to no biogenic material or ice-rafted debris (Supplementary Fig. 2a–c) occur as discrete unitsand laminations in sediment cores across the western Ross Sea(Fig. 1a and Supplementary Table 1). Given a shared 10 µm grain-size mode, these deposits were sourced directly from subglacialtill (Supplementary Fig. 2d–f). We interpret these as meltwaterdeposits that were transported by subglacial channels to thegrounding line and then dispersed in the ocean before settling tothe seafloor. Interestingly, the deposits are identical to meltwater‘plumite’ deposits in the Amundsen Sea41 (Supplementary Fig. 2a)and those described beneath the contemporary Pine Island Glacierice shelf42. In the western Ross Sea, the deposits are confinedto proximal grounding line and open marine units, supportingsediment deposition from buoyantmeltwater plumes expelled at thegrounding line. It is likely that these plumes contributed to meltingof ice at the grounding line and to erosion of the ice shelf base.

Upstream subglacial lakesWithin the upstream segment of the subglacial hydrological system(Fig. 1b), we identify a series of subglacial lakes that were dammedby grounded ice (Fig. 4a,b) when the middle and downstreamsegments of the hydrological system were active. An acoustic profilefrom this area shows laminated deposits that fill a small basin,stratigraphically bounded by basement rocks and post-LGM openmarine deposits and filled with subglacial lake deposits (Fig. 4c,d).Grounding-line landforms surround the basins, but are absentwithin the basins themselves (Figs 3a and 4a), and indicate anothermajor shift in retreat direction from east to west across the channelto south to north across the subglacial lakes (Figs 1b and 3a).This shows that the grounding-line retreat continued to follow theestablished subglacial hydrological system.

694

© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.

NATURE GEOSCIENCE | VOL 10 | SEPTEMBER 2017 | www.nature.com/naturegeoscience

NATURE GEOSCIENCE DOI: 10.1038/NGEO3012 ARTICLESa

Drainage frequency inferredfrom grounding-line landforms

SLW48

0 2 4 6 8 10 12 14 16 18 200

200

400

600

80010,000 km3 catchment1,300 km3 lake area

Lake

fill

time

(yr)

Basal melt rate (mm yr−1)

Geothermal heat (mW m−2)

0 100 200 300 400 500 600

Geothermal heat (mW m−2)

0 100 200 300 400 500 600

Ice

thic

knes

s, H

(m)

Ice

thic

knes

s, H

(m)

500

700

900

1,100

1,300

1,500Melt rate = 0.5 mm yr−1 Melt rate = 3 mm yr−1

us = 100 m yr−1, αs = −0.0001, bn = 100 mm yr−1

us = 100 m yr−1, αs = −0.0001, bn = 300 mm yr−1

us = 100 m yr−1, αs = −0.0005, bn = 100 mm yr−1

us = 100 m yr−1, αs = −0.0005, bn = 300 mm yr−1

us = 1,000 m yr−1, αs = −0.0001, bn = 100 mm yr−1

us = 1,000 m yr−1, αs = −0.0001, bn = 300 mm yr−1

us = 1,000 m yr−1, αs = −0.0005, bn = 100 mm yr−1

us = 1,000 m yr−1, αs = −0.0005, bn = 300 mm yr−1

SLW48 SLW48

500

700

900

1,100

1,300

1,500b c

Figure 5 | Conditions required to fill subglacial lakes. a, Timing of lake filling with a range of melt rates across the larger catchment and smaller subglaicallake area, using a 2 km3 lake volume. b,c, Geothermal heat flux required to produce melt rates of 0.5 and 3 mm yr−1, respectively, under a range ofconditions for ice thickness (H), accumulation rates (bn), and surface slopes (αs) and velocities (us). Lower (higher) basal melt rates would be achieved byshifting the curves to the left (right) by∼10 mW m−2 per 1 mm yr−1. Grey dashed line shows modelled palaeo-ice thickness over the subglacial lakes(Methods). Subglacial Lake Whillans, SLW.

The individual lakes have depths of 4–20m and volumecapacities of 0.002–0.8 km3 (Fig. 4b); the combined volume capacityis approximately 2 km3. Althoughprominent basement highs appearto separate some of the lakes (Fig. 4e), meltwater probably flowedfrom upstream to downstream subglacial lakes via hydropotentiallows and depressions in basement relief (Fig. 4b). This is supportedby the presence of meltwater fan deposits on the slope of a basementhigh at the upstream end of a subglacial lake (Fig. 4a,b), indicatingnorth to south meltwater flow. We suggest that the lakes ultimatelydrained into the channel system along the southern sill of the largestlake (Figs 3a and 4a,b). If the lakes completely and simultaneouslydrained, individual channelized drainage events would have lastedabout half a year with our calculated channel discharge, or shorterif the lakes did not drain completely or simultaneously.

To what extent the lakes were fed by upstream sources is un-clear, as there are no observed channels upstream of the lakes,and distributed drainage is difficult to identify in the geologicrecord. Distributed sources account for >90% of meltwater storedwithin contemporary subglacial lakes of the MacAyeal Ice Stream43

(Fig. 1a). However, it is likely that the drainage catchment for thewestern Ross Sea lakes was small (<10,000 km2), since the lakes areperched on the flank of a bank that was a sustained hydropotentialhigh while ice was grounded in this area (Supplementary Fig. 1).If we assume melt across the whole catchment reached and filledthese lakes through a distributed drainage system, basal melt ratesof >0.5mmyr−1 would be sufficient to produce a lake fill timeconsistent with the drainage event frequency of at most severalcenturies inferred from grounding-line landforms (Fig. 5a). Con-sidering the potential absence of significant upstream distributed

meltwater sources, an in situ basal melt rate of >3mmyr−1 overthe area of the subglacial lakes (1,300 km2; see Fig. 4b) would alsoproduce lake filling on the same timescale (Fig. 5a).

Such rates, under reasonable ice sheet conditions (ice flowvelocity, surface slope, accumulation), demand aminimumgeother-mal heat flux of 90–120mWm−2 (Fig. 5b,c and SupplementaryMethods). The subglacial hydrological system is in a region of lateCenozoic rifting44 and proximal to Quaternary volcanic islands andseamounts45 (Fig. 1a–b), where measurements of geothermal heatfluxes span from∼60–165mWm−2 (refs 46,47). Along the same riftzone, a heat flux of 285± 80mWm−2 was measured at the base ofSubglacial Lake Whillans, near the Whillans Ice Stream groundingline (Fig. 1a), with a basal melt rate of ∼18mmyr−1 (ref. 48).This melt rate is considerably higher than rates required to fill thewestern Ross Sea lakes. Therefore, we suggest that elevated, butgeologically feasible, geothermal heat fluxes could have producedsufficient basal melting either across the ice stream catchment orconfined to the area of the subglacial lakes over periods of decadesto several centuries.

ConclusionsAn extensive sediment-based subglacial channel network was reac-tivated numerous times during the post-LGM deglaciation of thewestern Ross Sea. Channelizedmeltwater drainage locally restrictedgrounding-line landform growth and, consequently, contributedto local grounding-line instability. Channel segments both bisectand are buried by grounding-line landforms, suggesting meltwaterdrainage events occurred episodically, and at periodicities of tens toseveral hundreds of years. The channel networkwas fed by upstream

NATURE GEOSCIENCE | VOL 10 | SEPTEMBER 2017 | www.nature.com/naturegeoscience

© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.

695

ARTICLES NATURE GEOSCIENCE DOI: 10.1038/NGEO3012

subglacial lakes in an area of geologically recent rifting, activevolcanism, and elevated geothermal heat flow. Meltwater drainageconfiguration appears to have persisted through various phases ofgrounding-line retreat, shifts in ice flow direction, and a circuitousretreat pattern, suggesting that the stable location of source lakes andample production of basal melting exerted a degree of influence onthe retreating ice stream.The probable recurrence of drainage eventsduring grounding-line landform construction suggests that an indi-vidual drainage event is not capable of dislodging a stable groundingline. It does, however, remain a possibility that repeated drainagethrough embayments, the development of pronounced grounding-line sinuosity and feedbacks with plume-driven ice melting, mayundermine grounding-line stability.

MethodsMethods, including statements of data availability and anyassociated accession codes and references, are available in theonline version of this paper.

Received 12 May 2017; accepted 21 July 2017;published online 21 August 2017

References1. Peters, L. E., Anandakrishnan, S., Alley, R. B. & Smith, A. M. Extensive storage

of basal meltwater in the onset region of a major West Antarctic ice stream.Geology 35, 251–254 (2007).

2. Perol, T., Rice, J. R., Platt, J. D. & Suckale, J. Subglacial hydrology and ice streammargin locations. J. Geophys. Res. 120, 1352–1368 (2015).

3. Alley, R. B., Blankenship, D. D., Bentley, C. R. & Rooney, S. Deformation of tillbeneath ice stream B, West Antarctica. Nature 322, 57–59 (1986).

4. Engelhardt, H. & Kamb, B. Basal hydraulic system of a West Antarctic icestream: constraints from borehole observations. J. Glaciol. 43, 207–230 (1997).

5. Palmer, S. J. et al . Greenland subglacial lakes detected by radar. Geophys. Res.Lett. 40, 6154–6159 (2013).

6. Howat, I. M., Porter, C., Noh, M. J., Smith, B. E. & Jeong, S. Sudden drainage ofa subglacial lake beneath the Greenland Ice Sheet. Cryosphere 9,103–108 (2015).

7. Wright, A. & Siegert, M. A fourth inventory of Antarctic subglacial lakes.Antarct. Sci. 24, 659–664 (2012).

8. Gray, L. et al . Evidence for subglacial water transport in the West Antarctic IceSheet through three-dimensional satellite radar interferometry. Geophys. Res.Lett. 32, L03501 (2005).

9. Wingham, D. J., Siegert, M. J., Shepherd, A. & Muir, A. S. Rapid dischargeconnects Antarctic subglacial lakes. Nature 440, 1033–1036 (2006).

10. Scambos, T. A., Berthier, E. & Shuman, C. A. The triggering of subglacial lakedrainage during rapid glacier drawdown: Crane Glacier, Antarctic Peninsula.Ann. Glaciol. 52, 74–82 (2011).

11. Fricker, H. A., Siegfried, M. R., Carter, S. P. & Scambos, T. A. A decade ofprogress in observing and modeling Antarctic subglacial water systems.Phil. Trans. R. Soc. 374, 20140294 (2016).

12. Bell, R. E., Studinger, M., Shuman, C. A., Fahnestock, M. A. & Joughin, I. Largesubglacial lakes in East Antarctica at the onset of fast-flowing ice streams.Nature 445, 904–907 (2007).

13. Stearns, L. A., Smith, B. E. & Hamilton, G. S. Increased flow speed on a largeEast Antarctic outlet glacier caused by subglacial floods. Nat. Geosci. 1,827–831 (2008).

14. Siegfried, M. R., Fricker, H. A., Carter, S. P. & Tulaczyk, S. Episodic ice velocityfluctuations triggered by a subglacial flood in West Antarctica. Geophys. Res.Lett. 43, 2640–2648 (2016).

15. Bartholomew, I. et al . Seasonal evolution of subglacial drainage andacceleration in a Greenland outlet glacier. Nat. Geosci. 3, 408–411 (2010).

16. Cowton, T. et al . Evolution of drainage system morphology at aland-terminating Greenlandic outlet glacier. J. Geophys. Res. 118, 29–41 (2013).

17. Andrews, L. C. et al . Direct observations of evolving subglacial drainagebeneath the Greenland Ice Sheet. Nature 514, 80–83 (2014).

18. Horgan, H. J. et al . Estuaries beneath ice sheets. Geology 41, 1159–1162 (2013).19. Horgan, H. J., Christianson, K., Jacobel, R. W., Anandakrishnan, S. &

Alley, R. B. Sediment deposition at the modern grounding zone of Whillans IceStream, West Antarctica. Geophys. Res. Lett. 40, 3934–3939 (2013).

20. Alley, R. B., Anandakrishnan, S., Dupont, T. K., Parizek, B. R. & Pollard, D.Effect of sedimentation on ice-sheet grounding-line stability. Science 315,1838–1841 (2007).

21. Christianson, K. et al . Basal conditions at the grounding zone of Whillans IceStream, West Antarctica, from ice-penetrating radar. J. Geophys. Res. 121,1954–1983 (2016).

22. Le Brocq, A. M. et al . Evidence from ice shelves for channelized meltwater flowbeneath the Antarctic Ice Sheet. Nat. Geosci. 6, 945–948 (2013).

23. Alley, K. E., Scambos, T. A., Siegfried, M. R. & Fricker, H. A. Impacts of warmwater on Antarctic ice shelf stability through basal channel formation.Nature Geosci. 9, 290–293 (2016).

24. Marsh, O. J. et al . High basal melting forming a channel at the grounding lineof Ross Ice Shelf, Antarctica. Geophys. Res. Lett. 43, 250–255 (2016).

25. Lowe, A. L. & Anderson, J. B. Evidence for abundant subglacial meltwaterbeneath the paleo-ice sheet in Pine Island Bay, Antarctica. J. Glaciol. 49,125–138 (2003).

26. Nitsche, F. O. et al . Paleo ice flow and subglacial meltwater dynamics in PineIsland Bay, West Antarctica. Cryosphere 7, 249–262 (2013).

27. Anderson, J. B. & Fretwell, L. O. Geomorphology of the onset area of apaleo-ice stream, Marguerite Bay, Antarctic Peninsula. Earth Surf. Process.Landf. 33, 503–512 (2008).

28. Domack, E. et al . Subglacial morphology and glacial evolution of the Palmerdeep outlet system, Antarctic Peninsula. Geomorphology 75, 125–142 (2006).

29. Campo, J., Wellner, J. S., Lavoie, C., Domack, E. & Yoo, K. C. Glacialgeomorphology of the northwestern Weddell Sea, Eastern Antarctic PeninsulaContinental Shelf: shifting ice flow patterns during deglaciation.Geomorphology 280, 89–107 (2017).

30. Wellner, J. S., Heroy, D. C. & Anderson, J. B. The death mask of the Antarcticice sheet: comparison of glacial geomorphic features across the continentalshelf. Geomorphology 75, 157–171 (2006).

31. Greenwood, S. L., Gyllencreutz, R., Jakobsson, M. & Anderson, J. B. Ice-flowswitching and East/West Antarctic Ice Sheet roles in glaciation of the WesternRoss Sea. Geol. Soc. Am. Bull. 124, 1736–1749 (2012).

32. Halberstadt, A. R. W., Simkins, L. M., Greenwood, S. L. & Anderson, J. B.Paleo-ice sheet behaviour: retreat scenarios and changing controls in the RossSea, Antarctica. Cryosphere 10, 1003–1020 (2016).

33. Batchelor, C. L. & Dowdeswell, J. A. Ice-sheet grounding-zone wedges (GZWs)on high-latitude continental margins.Mar. Geol. 363, 65–92 (2015).

34. Bouvier, V., Johnson, M. D. & Påsse, T. Distribution, genesis and annual-originof De Geer moraines in Sweden: insights revealed by LiDAR. GFF 137,319–333 (2015).

35. Anderson, J. B. Antarctic Marine Geology (Cambridge Univ. Press, 1999).36. Dowdeswell, J. A., Ottesen, D., Evans, J., Cofaigh, C.Ó. & Anderson, J. B.

Submarine glacial landforms and rates of ice-stream collapse. Geology 36,819–822 (2008).

37. McMullen, K. et al . Glacial morphology and sediment formation in the MertzTrough, East Antarctica. Palaeogeogr. Palaeoclimatol. Palaeoecol. 231,169–180 (2006).

38. Bart, P. J. & Owolana, B. On the duration of West Antarctic Ice Sheet groundingevents in Ross Sea during the Quaternary. Quat. Sci. Rev. 47, 101–115 (2012).

39. Smith, B. E., Fricker, H. A., Joughin, I. R. & Tulaczyk, S. An inventory of activesubglacial lakes in Antarctica detected by ICESat (2003–2008). J. Glaciol. 55,573–595 (2009).

40. Carter, S. P. & Fricker, H. A. The supply of subglacial meltwater to thegrounding line of the Siple Coast, West Antarctica. Ann. of Glaciol. 53,267–280 (2012).

41. Witus, A. E. et al . Meltwater intensive glacial retreat in polar environments andinvestigation of associated sediments: example from Pine Island Bay, WestAntarctica. Quat. Sci. Rev. 85, 99–118 (2014).

42. Smith, J. A. et al . Sub-ice-shelf sediments record history of twentieth-centuryretreat of Pine Island Glacier. Nature 541, 77–80 (2017).

43. Carter, S. P. et al . Modeling 5 years of subglacial lake activity in the MacAyealIce Stream (Antarctica) catchment through assimilation of ICESat laseraltimetry. J. Glaciol. 57, 1098–1112 (2011).

44. Cooper, A. K., Davey, F. J. & Behrendt, J. C. The Antarctic Continental Margin:Geology and Geophysics of the Western Ross Sea Vol. 5B9, 27–65 (Earth ScienceSeries, Circum-Pacific Council for Energy and Mineral Resources, 1987).

45. Rilling, S., Mukasa, S., Wilson, T., Lawver, L. & Hall, C. New determinations of40Ar/39Ar isotopic ages and flow volumes for Cenozoic volcanism in the TerrorRift, Ross Sea, Antarctica. J. Geophys. Res. 114, B12207 (2009).

46. Blackman, D. K., Von Herzen, R. P. & Lawver, L. A. The Antarctic ContinentalMargin: Geology and Geophysics of the Western Ross Sea Vol. 5B, 179–189(Earth Science Series, Circum-Pacific Council for Energy and MineralResources, 1987).

47. Morin, R. H. et al . Heat flow and hydrologic characteristics at the AND-1Bborehole, ANDRILL McMurdo Ice Shelf Project, Antarctica. Geosphere 6,370–378 (2010).

48. Fisher, A. T., Mankoff, K. D., Tulaczyk, S. M., Tyler, S. W. & Foley, N. Highgeothermal heat flux measured below the West Antarctic Ice Sheet. Sci. Adv. 1,1500093 (2015).

696

© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.

NATURE GEOSCIENCE | VOL 10 | SEPTEMBER 2017 | www.nature.com/naturegeoscience

NATURE GEOSCIENCE DOI: 10.1038/NGEO3012 ARTICLES49. Arndt, J. E. et al . The international bathymetric chart of the Southern Ocean

(IBCSO) Version 1.0—a new bathymetric compilation coveringcircum-Antarctic waters. Geophys. Res. Lett. 40, 3111–3117 (2013).

50. Fretwell, P. et al . Bedmap2: improved ice bed, surface and thickness datasets forAntarctica. Cryosphere 7, 375–393 (2013).

AcknowledgementsThe authors thank J. Walder, S. Carter, C. Clark and T. Swanson for productivediscussions; B. Demet for assistance in data collection; and A. Fonseca,S. Rezvanbehbahani and J. Aroom for assisting with analyses. This project was supportedby the National Science Foundation (NSF-PLR 1246353, J.B.A.) and the SwedishResearch Council (D0567301, S.L.G.).

Author contributionsL.M.S. and J.B.A. conceived the research. L.M.S., J.B.A., S.L.G., L.O.P. andA.R.W.H. interpreted the geophysical data. L.M.S. and A.R.W.H. calculated channel

flow properties. L.A.S. calculated hydraulic pressure. L.M.S. and L.O.P. analysedsediment samples. D.P. and R.M.D. provided ice sheet model data. H.M.G. completedthe heat flow model. L.M.S. wrote the manuscript, with major contributions fromJ.B.A., S.L.G. and H.M.G. All authors read and commented onthe manuscript.

Additional informationSupplementary information is available in the online version of the paper. Reprints andpermissions information is available online at www.nature.com/reprints. Publisher’s note:Springer Nature remains neutral with regard to jurisdictional claims in published mapsand institutional affiliations. Correspondence and requests for materials should beaddressed to L.M.S.

Competing financial interestsThe authors declare no competing financial interests.

NATURE GEOSCIENCE | VOL 10 | SEPTEMBER 2017 | www.nature.com/naturegeoscience

© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.

697

ARTICLES NATURE GEOSCIENCE DOI: 10.1038/NGEO3012

MethodsBathymetry and acoustic stratigraphy.Multibeam bathymetry was collected oncruise NBP1502A aboard the RVIB Nathaniel B. Palmer using a KongsbergEM-122 system in dual swath mode with a 1◦×1◦ array and 12 kHz frequency.Vertical resolution varies from 0.07 to 0.2% of the water depth51 and horizontalresolution is approximately 0.02% the water depth. Data are gridded at 20m.Sub-bottom acoustic data were collected with a Knudsen CHIRP 3260 systemduring cruise NBP1502A using a frequency of 3.5 kHz and a 0.25ms pulse width.Two-way travel time was converted to depth using a velocity of 1,500m s−1.

Grounding-zone wedge formation time. The sediment volume of GZW 1 (Fig. 2a)of 3× 109 m3 or 9× 104 m3 m−1 (metre width) is based on landform area andthickness, measured from multibeam bathymetry and acoustic profiles. Two-waytravel times through sediment were converted to depth using an acoustic velocitybracketed by 1,500–1,750m s−1 (refs 52–54). A lower grounding-line sediment fluxof 200m3 yr−1 m−1 (metre width) is based on refs 4,38,55 and a higher flux of1,000m3 m−1 yr−1 is from ref. 51.

Meltwater channel flow calculations.Meltwater flow velocity and discharge werecalculated following refs 56,57 and using the smallest measured channeldimensions (Fig. 2d). Hydraulic potential across the channels is based onisostasy-corrected bed and ice surface elevations from an ice sheet model for amodel time slice at 20,000 years ago, representative of the LGM ice sheetconfiguration (Supplementary Fig. 1). Because the ice sheet model timing is poorlyconstrained during the deglaciation of the continental shelf, we chose to use thebest observationally constrained model time slice (at 20,000 years ago) to estimatehydropotential and, thus, channel flow properties that we advise readers to treat asorder-of-magnitude estimates. We did, however, evaluate hydropotential surfacesusing model outputs from deglacial time slices, and the hydropotential high onCrary Bank remains and hydropotential across the channel increases, which wouldlead to higher channel flow discharge yet still within the same order of magnitude.Stream bed roughness was estimated with a dimensionless frictionparameterization f for turbulent pipe flow58 by using equations (1) and (2) tosimultaneously solve for f and u (depth-averaged flow velocity).

f =8Φhρwu2

(1)

f −0.5=−1.8 log10

[( R3.7

)1.11

+6.9

ρwuh/v

](2)

whereΦ is hydraulic potential gradient, h is the channel depth, ρw is the density offresh water, v is kinematic viscosity of water, and R is relative roughness

R=D90

h(3)

in which grain-size diameter at 90% of the distribution (D90; 1.39× 10−4 m) wasused as an estimate for equivalent roughness, based on grain-size measurementsfrom Ross Sea till matrix material (Supplementary Fig. 2d). Although till containslarger clasts (for example, granules and pebbles), we assume the abundantfine-grained matrix is more representative for calculating relative roughness at thebase of the channel. A value of 0.097 was used for f, which is similar to thecommonly applied value of 0.1 (refs 56,57). Discharge was calculated from theresulting flow velocity and a triangular channel cross-sectional area, assumingbank-full conditions and a flat ice base. Lateral variations in meltwater flowvelocity due to the channel banks were not considered, as the channels are muchwider than deep, and lateral effects are assumed to be minimal.

Ice sheet model. The ice sheet model is described by refs 59–61. It uses hybrid icedynamical equations with parameterized flux at the grounding line, allowing the

realistic simulation of floating ice shelves, ice streams and grounding-linemigration in long-term runs. Elevation surfaces are corrected for glacial isostaticadjustment. The simulation used here is from an intermediate model version,similar to that in ref. 60, with the addition of sub-ice ocean meltingparameterization depending on proximal 400m ocean temperatures61, which inthis run were obtained from archived output of the coupled-GCM experiment ofref. 62. The model was run transiently from 50,000 years ago to the present, on a20 km grid that spans all of Antarctica.

Sedimentological properties. Sediment samples were sieved prior to measurementwith a 500-µm sieve and grain sizes were measured with a Malvern 2000 grain-sizeanalyser. The till measurements do not include the coarser grains (granules andpebbles) and represent the matrix grain size. Core details and sample depths aresummarized in Supplementary Table 1.

Heat flow model. The conditions required for basal melting were assessed for arange of parameters using a one-dimensional model similar to that of ref. 63.Further details are in the Supplementary Methods.

Data availability. The authors declare that the data supporting the findings of thisstudy are available within the article and its Supplementary Information file.Multibeam bathymetry data used in this study collected aboard cruise NBP1502Aare available in an XYZ format from the corresponding author upon request.Sediment samples from cores used in the study are available by request from theAntarctic Core Collection at the Oregon State University Marine and GeologyRepository. Raw sedimentological data is available from the corresponding authorupon request.

References51. Jakobsson, M. et al . Geological record of ice shelf break-up and grounding line

retreat, Pine Island Bay, West Antarctica. Geology 39, 691–694 (2011).52. O’Regan, M. et al . Constraints on the Pleistocene chronology of sediments

from the Lomonosov Ridge. Paleoceanography 23, PA1S19 (2008).53. Anderson, J. B., Jakobsson, M. & Party, O. S. Oden Southern Ocean 0910

OSO0910 Cruise Report (University of Stockholm, 2010).54. Klages, J. P. et al . Retreat of the West Antarctic Ice Sheet from the western

Amundsen Sea shelf at a pre-or early LGM stage. Quat. Sci. Rev. 91,1–15 (2014).

55. Anandakrishnan, S., Catania, G. A., Alley, R. B. & Horgan, H. J. Discovery oftill deposition at the grounding line of Whillans Ice Stream. Science 315,1835–1838 (2007).

56. Walder, J. S. & Fowler, A. Channelized subglacial drainage over a deformablebed. J. Glaciol. 40, 3–15 (1994).

57. Ng, F. S. Canals under sediment-based ice sheets. Ann. Glaciol. 30,146–152 (2000).

58. Haaland, S. E. Simple and explicit formulas for the friction factor in turbulentpipe flow. J. Fluids Eng. 105, 89–90 (1983).

59. Pollard, D. & DeConto, R. M. Modelling West Antarctic ice sheet growth andcollapse through the past five million years. Nature 458, 329–332 (2009).

60. Pollard, D. & DeConto, R. M. Description of a hybrid ice sheet-shelf model,and application to Antarctica. Geosci. Mod. Dev. 5, 1273–1295 (2012).

61. Pollard, D., Chang, W., Haran, M., Applegate, P. & DeConto, R. M. Largeensemble modeling of last deglacial retreat of the West Antarctic Ice Sheet:comparison of simple and advanced statistical techniques. Geosci. Mod. Dev. 9,1697–1723 (2016).

62. Liu, Z. et al . Transient simulation of last deglaciation with a new mechanismfor Bolling–Allerød warming. Science 325, 310–314 (2009).

63. Budd, W. F., Jenssen, D. & Radok, U. Derived Physical Characteristics of theAntarctic Ice Sheet (Australian National Antarctic Expeditions Interim Reports.Series A, Glaciology, no. 120; University of Melbourne, 1971).

© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.

NATURE GEOSCIENCE | www.nature.com/naturegeoscience